use cpu

tools.py

@@ -2257,7 +2257,7 @@ def create_llm_pipeline():
     return pipeline(
         "text-generation",
         model=model_id,
-        device_map="auto",
+        device_map="cpu",
         torch_dtype=torch.float16,
         max_new_tokens=1024,
         temperature=0.3,
@@ -3062,8 +3062,8 @@ Current date: {current_date_str}
 
     # Cleanup
     if result.get("done"):
-        torch.cuda.empty_cache()
-        torch.cuda.ipc_collect()
+        #torch.cuda.empty_cache()
+        #torch.cuda.ipc_collect()
         gc.collect()
         print("🧹 Released GPU memory after completion")